#Bug report from quick Sora test

4 messages · Page 1 of 1 (latest)

barren crescent
#

Model: openai/sora-2-pro
Job ID: T370UuaYKsYC7C3PoVlM
Generation ID: gen-vid-1774468176-JVusj78IJ3O2EjUa3djW
Requested Duration: 4s
I've been testing the API integration with my agent and ran into three specific issues regarding endpoint stability, documentation accuracy, and pricing mismatches.

Bug 1: The /content endpoint throws 404s (Doc Mismatch) The official API spec states that to download the video, we should run GET /api/alpha/videos/:jobId/content.

  • Actual Behavior: Hitting that endpoint directly returns a 404 Not Found. * Workaround Used: I had to parse the unsigned_urls array from the completed status JSON and use the specific string provided (.../content?index=0).
  • Request: Either the documentation should be updated to clarify that users must extract the URL from the unsigned_urls array, or the bare /content endpoint needs to actually route to the file.
#

Bug 2: Intermittent 404 errors while polling for status
While polling GET /api/alpha/videos/:jobId every 30 seconds for job status, the endpoint completely disappeared twice.

  • Flow: pending -> 404 Not Found -> 404 Not Found -> completed
  • Impact: Any strict client wrapper will crash assuming the job was deleted. Polling logic has to be manually written to swallow and ignore 404s during the generation window.
#

Bug 3: Overcharge on billing
The generation was requested at 4 seconds. According to the stated pricing (https://openai.com/api/pricing/) of ~$0.10/second for Sora 2 on this page, this should have cost around $0.40 .

  • Actual Charge: The API returned "usage":{"cost":1.188,"is_byok":false}. * Impact: The hold and eventual charge was nearly 3x the expected price for a standard 4-second render.
Pricing | OpenAI API

Pricing information for the OpenAI platform.

true night
#

Thank you for the detailed reports!

I could not reproduce the behavior you're talking about in #1 - I am seeing the /content return the file, with or without the index param as long as I call it once the generation is complete. Are you still seeing this?

#2 - We're working on a fix for this right now

#3 - The billing is correct, but I think I made a mistake in the pricing chart I linked from the docs. It was showing the costs for Sora, when the model we support is Sora Pro. I've updated the links and you can find the actual pricing here: https://developers.openai.com/api/docs/pricing#video